Caching

Caching can reduce the cost of repeated input. Support, retention and prices depend on the selected provider. The rule below applies only to Chat routes with 20-minute and 120-minute write rates.

For a Chat route listing 20-minute and 120-minute cache-write prices, you may pass numeric cache_ttl in JSON or one numeric X-Cache-TTL header. Both values are forwarded unchanged to the selected provider.

Omitting both selects 20 minutes. If either value is 120 or greater, the 120-minute tariff applies; otherwise the 20-minute tariff applies. This also resolves conflicting body and header values.

For such a route, an invalid or repeated header or a nonnumeric cache_ttl returns 400 before provider dispatch.

Under the Vokis tariff, the selected write price applies to all input tokens not read from cache, including new tokens after a cached prefix. The read price applies to cached input tokens. A write-tariff charge does not claim that the provider reported actual cache creation.

Other routes retain their own cache and billing rules; check the selected provider prices in the model catalog.